Touted as the next frontier of artificial intelligence, agentic AI hogged the data management headlines in 2025.
Seemingly ushering the realisation of the no-more-drudge-work predictions that heralded the arrival of general AI years back, agentic AI has certainly become the target of institutional investment and developer innovation in the past 12 months.
According to a recent Gartner report, AI agents will be behind more than half of all business decisions by 2027, and a separate survey for Soundhound AI found that seven in 10 US bank executives said they expect the technology to have a “significant” impact on their business.Meanwhile, the market for AI agents within the financial sphere has been forecast to grow to $80 billion within a decade from $2.1 billion in 2024, according to a report in the Financial Times.
Turbo-charged Gains
That investment is expected to bring substantial productivity and efficiency gains. McKinsey analysis estimates that scaled agent deployments could bring productivity improvements of three-to-five per cent annually, potentially lifting growth by 10 per cent. That would reverse what the consultancy said has been damp-squib adoption of AI more generally, with eight out of 10 companies across the economy deriving no benefit from general AI at all.
AI’s transformation from co-pilot to autopilot is bringing near-autonomous capabilities to workflows across financial institutions. Customised to perform specific tasks, agents are being deployed to automatically carry out different functions without the need for continual prompts and suggestions.It’s seen as bringing streamlined data-management capabilities, analytical heft and collaborative innovation among a host of other technological gains.
“An agentic AI model is one that anticipates needs, suggests actions, and makes decisions proactively in alignment with defined prompts or objectives,” said Rinesh Patel, Global Head of Industry, Financial Services at AI data cloud company Snowflake.
Kalyan Kumar chief product officer at HCLSoftware places his focus on the technology’s ability to “learn”.
“Agentic systems actually combine experience of interaction, bring in contextual data and can also perform business operations or IT operations,” Kumar told Data Management Insight. “It adapts.”
Partner, Not Tool
As data processes become more complex and more management tools become available, firms have begun looking for ways to automate those tasks in an “always-on” way. That can potentially bring huge benefits, Patel added.
“The way organisations are now thinking about agentic models is you think about them more as a partner as opposed to a tool. You can quickly see the benefits of that if you think about the efficiency gains, the productivity gains, the things that we’re doing in daily life, and some of the the low IP work that takes place, and how agentic models could potentially support our day-to-day workflow.”
One of the core characteristics that sets agentic AI from other robotic automation systems is its ability to create its own rules in pursuit of its prescribed tasks. It can hold memory and understand intent. As well, it can be organised hierarchically to manage the most complex of operations, the so called router agents.
Levent Ergin, chief climate, sustainability and AI strategist at Informatica, explained the importance of this characteristic by likening it to the dynamics of an orchestra.
Orchestrator agents are “the conductor who orchestrates” the workflow, Ergin told Data Management Insight. “And the job would be broken up by a planner agent, and you’d have another head that’s controlling the violins, controlling the drums, controlling all the other aspects.”
In his analogy, planner agents are composers or arrangers, making sure the workflow is properly conceptualised. Beneath those are style agents that would lead individual instrument sections and executor agents, which are the musicians.
“They have a very specific boundary for an input, a process and an output,” he said.
Within such a structure, agents would be in constant communication through feedback loops.
“Based on those feedback loops the ecosystem is able to come up with its own decisions,” he said.
Product Launches
The appeal of agentic AI is reflected in the flurry of new products that have been launched by vendors and software developers, both new and established, in the past year. The variety of agents illustrated the broad range of use cases to which it is being applied.
Informatica enhanced its agentic offering with an update to CLAIRE AI platform in the spring, providing orchestration capabilities for a variety of agents within its data and metadata management services.
In the regulatory space, Solidus Labs evolved its Halo agentic-based platform to offer end-to-end risk visibility for compliance teams.
In the summer, data observability provider Acceldata unveiled its Agentic Data Management (ADM) platform that its said marked “a transformative leap from data observability—unifying governance, quality, catalog, and observability into an intelligent, agent-driven system”.
And among start-ups, Wand.ai expanded its own agentic offering with a “multi-agent cognitive layer” that acts as an intermediary between clients and its AI models.
Regulatory Benefits
One of the growth centres of agentic applications in compliance where agents are seen as useful analysing and triaging the many thousands of alerts that teams receive to eliminate false positives. It’s a process that would normally take hours to achieve manually but can be completed in minute with agents, said Chen Arad, co-founder of trade surveillance technology provider Solidus Labs.
“We’re seeing active adoption of agentic AI for… the remediation, the compliance workflows, the false positive resolution, etc that’s happening now… a process that could take more than an hour can be done in five minutes,” Arad told Data Management Insight.
Informatica’s Ergin said agentic AI could be applied to the long-running struggles banks have had in complying with BCBS 239. The Basel Committee on Banking Supervision’s Principles for effective risk data aggregation and risk reporting was introduce in 2013 and requires institutions to take steps to secure and ensure optimal quality of their risk data capabilities and risk reporting.
Data management hurdles, talent shortages and poor succession planning are bedevilling banks as they struggle to respond to the directive.
“You can actually point the agentic AI technology to a lot of those broken areas where you can actually significantly improve your controls around the data that you have,” Ergin said. “And that’s certainly one of the areas that we are seeing agentic AI being piloted.”
Solidus’ Arad, agreed and also pointed to market surveillance detection as a target for agentic applications, in regular and crypto markets.
Cautious View
While the potential and apparent benefits of agentic AI have been debated intensively, co have the downsides, real and theoretical.
Agents are expensive to build and deploy and the jury is out on when firms can expect a return on their investment into these structures. As Ergin wrote in an editorial for Data Management Insight, 35 per cent of firms are struggling to secure C-suite backing because they’re unable to demonstrate the value of their pilot projects.
That links into another consideration that is making adoption tricky: cultural acceptance. Change is had to achieve if the professionals trusted with brining it are suspicious that it will lose them their jobs, change their status or make their work-life more burdensome.
But the biggest question that has loomed over agentic AI since its first iterations is a theoretical and ethical one: how much autonomy should agents be granted? At a recent Bloomberg Enterprise Technology and Data Summit in London, experts from capital markets mulled the risks of handing further decision-making control to agents.
It was agreed that top-level authorisations should still be reserved for professionals and that humans still had to be involved in checking the outputs of lower-level processes. That despite predictions that within six months, agents will be as adept at making decisions as humans.
“We’re a couple of years into this AI journey; it’s quite new and I think that there is an element of caution that you want to apply still for the things that are really risk taking decisions,” Head of Data and Machine Learning, Man Group, told delegates. “We need to get the governance framework right there.”
Subscribe to our newsletter
